What is a major benefit of serverless architecture in AI applications?

Reduces infrastructure management overhead

What is a key limitation of automated feature engineering tools?

They sometimes generate redundant or irrelevant features

What is a primary goal of Qwen-2 by Alibaba?

Expanding multilingual capabilities across 27 languages

What is the primary advantage of AutoML?

It automates model selection, feature engineering, and hyperparameter tun

What is the purpose of Explainable AI (XAI)?

Making AI decisions more transparent and understandable

What is the main advantage of multi-query attention (MQA) in LLMs?

Faster inference and decoding speed

In real-time analytics with Edge AI, why is it important to process data closer to the source?

It decreases latency and improves real-time decision-making

What optimization in Falcon 180B improves its efficiency over previous models?

Multi-query attention (MQA) for faster decoding

Which is a major challenge of AI regulation?

Stricter laws on AI-generated content and data privacy compliance

The bitsandbytes library enables:

Stable 8-bit and 4-bit quantization of large AI models

The main reason AI APIs often use serverless architectures is:

It allows automatic scaling without provisioning fixed resources

Which of the following is an example of structured data?

A relational database storing customer transactions

A diagram in the slides explains model drift as:

A decrease in model performance over time due to changing data patterns

What is feature engineering?

The process of creating new features from existing ones in a dataset

The importance of continuous retraining in MLOps is best justified by:

The need for adapting models to new data distributions

What is the purpose of LoRA (Low-Rank Adaptation) in training LLMs?

It allows fine-tuning with minimal additional parameters

What is the difference between data and information?

Data is raw facts, while information has context

What is Federated Learning?

A decentralized method of training models without sharing raw data

What is a key improvement of LLaMA-2 over LLaMA-1?

LLaMA-2 increased the number of training tokens + commercial license

What is a major limitation of third-party AI APIs?

Data privacy concerns due to cloud-based processing

What is the main advantage of vector databases for AI applications?

They enable fast similarity search for high-dimensional embeddings

studbud

Emerging Topics Test 1

25 termsStudBud Administration102 plays

Imported from external platform

Flashcards25 cardsFPractice quizPHost liveL

Questions 25

Answers

1
What is a major benefit of serverless architecture in AI applications?
Increases computational costs
Only works with structured data
Requires manual scaling
Reduces infrastructure management overhead
2
What is a key limitation of automated feature engineering tools?
They require expensive GPUs to run
They cannot handle missing values
They do not support structured data
They sometimes generate redundant or irrelevant features
3
What is a primary goal of Qwen-2 by Alibaba?
Creating an LLM exclusively for self-driving applications
Reducing model size to fit into consumer GPUs
Removing transformer-based architecture from AI models
Expanding multilingual capabilities across 27 languages
4
Which LLM model is based on a Sparse Mixture of Experts (SMoE)?
Falcon
Mixtral
Phi-2
LLaMA-3
5
What is the primary advantage of AutoML?
It guarantees 100% accuracy in predictions
It eliminates the need for machine learning expertise
It automates model selection, feature engineering, and hyperparameter tun
It removes the need for data preprocessing
6
What is the purpose of Explainable AI (XAI)?
Increasing AI model complexity
Improving dataset storage efficiency
Eliminating bias in AI models
Making AI decisions more transparent and understandable
7
What is the main advantage of multi-query attention (MQA) in LLMs?
Faster inference and decoding speed
Elimination of the need for training
Reduced dataset requirements
Increased token count per batch
8
In real-time analytics with Edge AI, why is it important to process data closer to the source?
It decreases latency and improves real-time decision-making
It eliminates the need for cloud servers
It simplifies database queries
It reduces data storage costs
9
What optimization in Falcon 180B improves its efficiency over previous models?
Integration of a physics-based reasoning module
Use of reinforcement learning instead of traditional fine-tuning
Multi-query attention (MQA) for faster decoding
Elimination of positional embeddings for better generalization
10
Which is a major challenge of AI regulation?
Stricter laws on AI-generated content and data privacy compliance
A complete ban on AI applications in enterprise settings
The removal of Explainable AI (XAI) requirements
Decreasing interest in ethical AI research
11
Which of the following is NOT a key component of MLOps?
Automated feature engineering
Model monitoring
Continuous Integration/Continuous Deployment (CI/CD)
Data versioning
12
The bitsandbytes library enables:
Faster inference at the cost of removing critical model layers
Stable 8-bit and 4-bit quantization of large AI models
Conversion of deep learning models to tabular datasets
The removal of all attention layers in an LLM
13
Which type of data includes JSON and XML formats?
Semi-structured
Unstructured
Structured
Binary
14
The main reason AI APIs often use serverless architectures is:
It increases the computational cost significantly
It ensures complete control over AI model training
It prevents cloud-based AI from requiring user authentication
It allows automatic scaling without provisioning fixed resources
15
Which of the following is an example of structured data?
A relational database storing customer transactions
A collection of tweets and Facebook posts
A dataset of random text files
A video recorded from a security camera
16
A diagram in the slides explains model drift as:
A decrease in model performance over time due to changing data patterns
A type of hyperparameter tuning
A data leakage issue caused by incorrect feature selection
A cybersecurity threat in AI models
17
What is feature engineering?
A way to eliminate redundant features from a dataset
A method to remove outliers from a dataset
The process of creating new features from existing ones in a dataset
A technique to convert all features into categorical variables
18
The importance of continuous retraining in MLOps is best justified by:
The need for adapting models to new data distributions
The requirement of reducing computation costs
The inability of models to generalize
The necessity of increasing dataset sizes
19
What is the purpose of LoRA (Low-Rank Adaptation) in training LLMs?
It eliminates the need for pre-training
It removes bias from datasets
It allows fine-tuning with minimal additional parameters
It replaces the original model’s weights completely
20
Which open-source tool is used for automatic feature engineering?
Scikit-learn
Pandas
TensorFlow
Featuretools
21
What is the difference between data and information?
Information is always more precise than data
Data is raw facts, while information has context
Data is only numerical, while information is textual
Data and information are the same
22
What is Federated Learning?
A cloud-based API for training AI models
A model compression technique
A reinforcement learning algorithm
A decentralized method of training models without sharing raw data
23
What is a key improvement of LLaMA-2 over LLaMA-1?
LLaMA-2 introduced reinforcement learning from human feedback (RLHF)
LLaMA-2 removed attention mechanisms to reduce computational costs
LLaMA-2 increased the number of training tokens + commercial license
LLaMA-2 switched from transformer architecture to RNNs
24
What is a major limitation of third-party AI APIs?
They are free to use
They require high local computational power
Data privacy concerns due to cloud-based processing
Lack of cloud integration
25
What is the main advantage of vector databases for AI applications?
They eliminate the need for feature engineering
They store structured tabular data more efficiently
They enable fast similarity search for high-dimensional embeddings
They are less expensive than relational databases

Emerging Topics Test 1 Flashcards | StudBud